Upgrade to Pro — share decks privately, control downloads, hide ads and more …

[CVPR20 Tutorial] Opening

[CVPR20 Tutorial] Opening

[CVPR20 Tutotrial] Image Retrieval in the Wild
Opening
Yusuke Matsui, Takuma Yamaguchi, and Zheng Wang
https://matsui528.github.io/cvpr2020_tutorial_retrieval/

Yusuke Matsui

June 16, 2020
Tweet

More Decks by Yusuke Matsui

Other Decks in Research

Transcript

  1. 1
    Image Retrieval in the Wild
    CVPR 2020 Tutorial on
    Yusuke Matsui
    The University of Tokyo
    Takuma Yamaguchi
    Mercari Inc.
    Zheng Wang
    National Institute of Informatics

    View full-size slide

  2. 2
    Yusuke Matsui
    ✓ Image retrieval
    ✓ Large-scale indexing
    http://yusukematsui.me
    “Reconfigurable Inverted Indexing”, ACMMM 18 “A Survey of Product Quantization”, ITE 18
    Lecturer (Assistant Professor), the University of Tokyo, Japan
    @utokyo_bunny

    View full-size slide

  3. 3
    Takuma Yamaguchi
    Engineering Manager, Mercari, Inc.
    https://www.linkedin.com/in/kumon-takuma-yamaguchi/
    • OCR
    • Image Restoration
    • Super Resolution
    • Distributed File/Computing System
    • Image Recognition System
    • Visual Search System

    View full-size slide

  4. 4
    Zheng Wang
    ✓ Person Re-identification
    ✓ Instance Search
    https://wangzwhu.github.io/home/
    “Learning to Reduce Dual-level Discrepancy for
    Infrared-Visible Person Re-identification”, CVPR 19
    “Beyond Intra-modality: A Survey of Heterogeneous
    Person Re-identification”, IJCAI 20
    JSPS Fellowship Researcher, National Institute of Informatics, Japan

    View full-size slide

  5. 5
    Image retrieval
    ➢ Significant progress has been made in the last decade
    ➢ However, evaluation is based on existing datasets
    ➢ No enough discussion about how to create a practical and large-
    scale visual search system
    Paris dataset: https://www.robots.ox.ac.uk/~vgg/data/parisbuildings/

    View full-size slide

  6. 6
    Practical and large-scale visual search system for
    real-world applications
    ✓ Discovering visually similar items in online marketplace apps
    ✓ Re-identification for pedestrians in a security scenario.

    View full-size slide

  7. 7
    Schedule
    ◆08:30 – 09:15 (Seattle) / 17:30 – 18:15 (Paris) / 24:30 – 25:15 (Tokyo)
    Billion-scale Approximate Nearest Neighbor Search - Yusuke Matsui
    ◆09:15 – 10:00 (Seattle) / 18:15 – 19:00 (Paris) / 25:15 – 26:00 (Tokyo)
    A Large-scale Visual Search System in the C2C Marketplace App Mercari - Takuma Yamaguchi
    ◆10:00 – 10:30 (Seattle) / 19:00 – 19:30 (Paris) / 26:00 – 26:30 (Tokyo)
    Coffee Break
    ◆10:30 – 11:15 (Seattle) / 19:30 – 20:15 (Paris) / 26:30 – 27:15 (Tokyo)
    Beyond Intra-modality Discrepancy: A Survey of Heterogeneous Person
    Re-identification - Zheng Wang
    ◆11:15 – 12:00 (Seattle) / 20:15 – 21:00 (Paris) / 27:15 – 28:00 (Tokyo)
    Live-coding Demo to Implement an Image Search Engine from Scratch - Yusuke Matsui

    View full-size slide